Local AI usage analytics

Track Claude Code and Codex token usage by project.

See tokens, approximate cost, cache efficiency, models, projects, and daily activity in one native Mac dashboard. Usage metadata stays on this Mac; prompts and responses are not included.

Download for macOS
Signed & notarized by Apple
Explore the dashboard
Claude CodeCodexLocal metadata only
One 30-day view across every configured local projectToday · 7 days · 30 days · All time
lpm Stats dashboard showing 30-day Claude Code and Codex token usage by day, project, provider, and recent session
Real local usage data shown in lpm. Cost is an estimate; Codex pricing is approximate.
Useful answers, not another counter

Know what your agents used—and why

The total matters. The provider, project, model, cache share, and session behind it matter more.

Understand where the tokens went

Separate fresh input, cached input, output, and reasoning tokens. See the peak day and an approximate model-aware cost beside the total.

Compare Claude Code and Codex

Read provider share at a glance, then switch the daily chart between token volume and percentage share to see how your workflow changes.

Trace usage back to the work

Rank projects by tokens or sessions and inspect recent runs with their provider, model, duration, recency, and token composition.

Every useful angle

From the whole month to one agent session

Start with the trend, then narrow the view until you know which project, provider, and model drove it.

Daily activity

See the rhythm, not just the total

Volume / Share
StartPeak day highlighted in your totalsNow

Provider share

Claude Code vs Codex

Claude CodeCodex

Token composition

Fresh, cached, output, reasoning

Check whether context is being reused efficiently and how much of the result comes from output or reasoning tokens.

Projects & sessions

Follow usage back to the codebase

Tokens · Sessions · Name
frontend42 sessions
api31 sessions
docs18 sessions

Open recent sessions to see provider, model, duration, recency, and the token mix behind each run.

Automatic by design

No spreadsheets. No manual tracking.

Your existing local agent histories already contain the usage metadata. lpm turns it into a useful project view.

  1. 01

    Run agents in a configured project

    Use Claude Code or Codex as usual. lpm matches each local session to the configured project root where it ran.

  2. 02

    Read local usage metadata

    lpm finds token and model metadata in the agents’ local session histories and groups it without adding prompts or responses to the dashboard.

  3. 03

    Explore the dashboard

    Switch from today to all time, compare providers, sort projects, and inspect recent sessions whenever you need context.

Your usage history stays yours.

The stats view is built from usage metadata in local Claude Code and Codex session histories. It gives you the operational picture without turning your work into another cloud analytics feed.

  • Usage metadata stays on this Mac
  • Prompts and responses are not included
  • Only configured local projects are counted
  • No hosted analytics account is required
FAQ

Claude Code and Codex token usage

What the dashboard counts, where the data comes from, and what stays private.

  • What can I track for Claude Code and Codex?
    lpm tracks total, input, cached input, output, and available reasoning-token metadata. You can break usage down by day, provider, project, model, and recent session, with Today, 7 days, 30 days, and All time ranges.
  • Where does lpm get the token usage data?
    lpm reads usage metadata from local Claude Code and Codex session histories, then matches each session’s working directory to a configured local lpm project.
  • Does the stats dashboard include my prompts or responses?
    No. Usage metadata stays on this Mac, and prompts and responses are not included in the stats dashboard.
  • Is the estimated cost the same as my bill?
    No. It is an estimate based on current public list prices per recognized model, with cached reads and writes priced separately. OpenAI and Codex pricing is approximate, so use your provider’s billing page as the source of truth.
  • Can I find which project or session used the most tokens?
    Yes. Sort projects by tokens or sessions, then inspect recent sessions with provider, model, duration, recency, and token composition details.
  • Are remote SSH projects included?
    Not currently. The stats dashboard counts configured local projects and excludes SSH projects.

Keep reading

Best terminal for Claude Code & Codex

Run AI coding agents in parallel while every service, terminal, and project stays in view.

Multiple Claude Code accounts

Pin work and personal Claude accounts to separate projects and run them side by side.

Connect agents to your dev environment

Give Claude Code and Codex tools to start services, read logs, and work across parallel copies.

Review AI agent changes

Inspect every file an agent changed in a native diff view before you commit.

Stop guessing where the tokens went.

Download lpm for macOS, run Claude Code and Codex in your configured projects, and turn local usage metadata into answers you can act on.

Download for macOS
Signed & notarized by Apple
View the source on GitHub